In New York, the AI Initial Establishment announced the completion of $320 million in financing, valued at $2.3 billion. This company wants to turn human operating data from video games into a training base for universal intelligence and to further use robotics, simulations and game scenes.
The company indicated that the cumulative total amount disclosed after the round had increased to $454 million. Prior to that, the company had completed $134 million in financing in October 2025.
Financing will be mainly for calculus.
General Information indicates that this round of financing was led by Khosla Ventures. The bulk of the new funds will be used to expand computing resources and advance the pre-training of the next generation of models.
The company has entered into cooperation with CoreWeave to continue to expand its training infrastructure. A further portion of the funds will be used to open the API, with the goal of increasing the number of clients to access by the end of the summer.
Game Operations Data is a core asset
The central idea of General Investment is not simply to let the model “see video”, but to let the model learn the specifics of the player in the game. In the company ' s view, key-clicking and action time-pointing labels help the model to understand the relationship between action and results better than a pure video.
According to the company, this type of data helps models to distinguish between “subjects” and “environment”, thus creating greater causal judgement. The goal is to train a unified model that allows it to make decisions in game images and to migrate to realistic devices such as robots.
Tested robotics and simulations
The company showed a four-foot robot and said that it could fine-tune the model with real robot data for only about eight minutes. The data were collected from the streets rather than from the office environment where the robots were located.
The company also demonstrated an in-house world model environment for training. Unlike traditional game engines, this type of environment is generated by model-by-frame. General Investment, which is seen as a training ground rather than a final product, is a model of smarts that can be deployed to different scenarios.
Commercialization targeting bottom model providers
Currently, the company has a small number of customers in the field of games, simulations and robotics. The management stated that in the future it wished to become a bottom model provider similar to OpenAI or Anthropic, allowing other enterprises to base their technology on automatic driving, industrial robotics or hazardous environmental operating systems.
In addition to modelling training, the company has recently launched a platform called Nerve, which provides players with data labelling, robotic remote operation, and so forth, in the hope of expanding data sources and creating closed loops from game users to model training to commercial deployment.
Additional information:Pim de Witte, the founder, stated that the company did not want its intelligence to be used to harm human beings, but could accept such uses as search and rescue.